PresentationPDF Available

Automated cloud-based workflow for quantification of MRI signal intensity - initial real-world clinical validation



One-third of brain MRI scans performed worldwide make use of gadolinium-based contrast agent injections to enable detection of the breakdown in the blood-brain barrier by the resulting enhancement. Observations have shown increased signal intensity, particularly in the globus pallidus and thalamus after multiple doses of linear contrast agents but there is no standard procedure to measure this contrast intensity. Current quantitative methods are manual, labor intensive, time-consuming and provide variable results. We present a fully automatic workflow which accelerates the investigation of signal intensity in these nuclei after multiple doses of contrast agents by extracting the T1-weighted modal intensity value and applying appropriate corrections and normalizations to allow comparison across acquisitions and protocols. Automatic results matched up to 94% correlation with manual results and reduced the time by 90%.
Automated cloud-based workflow for
quantification of MRI signal intensity:
initial real-world clinical validation
Marc Ramos 1, Vesna Prčkovska 1, Paulo Rodrigues 1, Jinnan Wang 2,
Franklin Moser 3, Markus Blank 2, Sheela Agarwal 2, Jacob Agris 2,
David Moreno-Dominguez 1
1 QMENTA Inc, 2 Bayer Radiology, 3 Cedars Sinai Medical Center
2019 - Marc Ramos - QMENTA - 2
Speaker Name: David Moreno-Dominguez
I have the following financial interest or relationship to disclose with
regard to the subject matter of this presentation:
Company Name: QMENTA
Type of Relationship: Employee and stock options holder
Declaration of
Financial Interests or Relationships
2019 - Marc Ramos - QMENTA -
Gadolinium contrast
Gadolinium (Gd) based contrast
agents (GBCA) have been widely
used in clinical MRI for the last 30
While GBCA has been considered
safe, Gd is highly toxic.
GBCA has been associated with
incidence of nephrogenic systemic
fibrosis, and recently, with
elevated signal intensities in
unenhanced T1-weighted images.
Standard intensity
Current standard procedure
consists of manual delineation of
brain regions of interest and
averaging of the signal intensity
(SI) within the regions.
It is labor-intensive,
time-consuming and susceptible
to rater bias.
Our approach
In this work we have developed a fully automated pipeline to reduce time and increase reproducibility,
and validated it against manual results 3
2019 - Marc Ramos - QMENTA -
Patients with at least eight sessions of GBCA-enhanced MRI scan.
Linear contrasts: Gadoversetamide, Gadobenate dimeglumine, Gadodiamide.
Macrocyclic contrasts: Gadobutrol.
Injected with only linear or macrocyclic.
113 patients with a total of 205 MRI sessions1
T1-weighted images from 1.5T and 3T Siemens scanners at Cedars-Sinai Medical Center.
1.5T: TR 1330 ms; TE 4.8 ms; TI 800 ms; flip angle 15°; section thickness 12.5 mm; matrix size: 256 × 192; echo-train length 1.
3T: TR 2100 ms; TE 3.0 ms; TI 900 ms; flip angle 9°; section thickness 11 mm; matrix size 256 × 256; echo train length 1.
1. Wang et al. (2018). Automated signal intensity quantification software:
initial “real world” clinical validation. In Western Neuroradiological Society 49th Annual Meeting.
DOI: 10.13140/RG.2.2.18057.90727
2019 - Marc Ramos - QMENTA -
Methods: Automatic Pipeline
1. Detect the most recent session (MRS) of patient with valid T1w image.
2. Atlas-based segmentation of ROIS in MRS image:
Globus Pallidus, Thalamus, Dentate Nucleus, Pons
3. Non-linear registration across timepoints using ANTs (to accommodate
region displacement due to deformations), warp MRS regions to all
longitudinal sessions.
4. Extract the SI modal values using maximum of Gaussian kernel-density
estimates (KDEs) curve.
5. Correction of SI values for differing sequence parameters using signal
equations and tissue constant values.
6. SI normalization using reference ROIs:
Pallidus-Thalamus; Dentate-Pons
2019 - Marc Ramos - QMENTA -
4. SI KDE and
modal value
1. Last
3. Non linear warping T1-w regions
2. Automatic
and SI Analysis
Input data - N sessions
6. SI correction
7. SI normalization
Methods: Automatic Pipeline
2019 - Marc Ramos - QMENTA -
SI Correction
Use of MR signal equation for each SI measurement, based on
mag. field strength, sequence type, and tissue type constants 1.
Allows for correction of differing TI, TR and TE parameters within
same sequence type.
Methods: SI Correction and Normalization
Equations for correcting by TE, TR and flip angle¹:
Inversion Recovery
1. Fletcher, Lynn M., John B. Barsotti, and Joseph P. Hornak. "A multispectral analysis of brain tissues." Magnetic Resonance in Medicine 29.5 (1993): 623-630.
SI Normalization
Globus pallidus values are normalized over thalamus ones.
Dentate nucleus values are normalized over pons ones.
Allows for comparison across scans
2019 - Marc Ramos - QMENTA -
Results: Initial clinical validation
Manual processing was performed by two radiologists who delineated the ROIs and recorded average
signal intensity.
Strong correlation was found between manual and automated analysis.
Very high correlation coefficients were found in all regions.
Dentate Nucleus: 0.94, Globus Pallidus: 0.9 and Pons: 0.93.
Manual results
Auto results
2019 - Marc Ramos - QMENTA -
Results: Automatic processing of all sessions
Session #
2019 - Marc Ramos - QMENTA -
Parallel cloud computing allowed the algorithm to process all sessions data in 10 hours, compared to 100 hours it
would take of radiologist time.
Manual results might be a biased gold-standard as the SI measurements during the manual review were made
based on oval ROI while the software segments whole structure.
Future work:
The tool can be easily extended to any other MRI sequence and ROI
Exploration of additional clinically relevant applications of the pipeline
Improve the correlation between manual vs. automatic values
Thank you for your
David Moreno-Dominguez
Booth 1010
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
A lot of research is now underway in trying to understand the mechanism of increased signal intensity (SI) and gadolinium presence in the brain and whether it has a clinical implication for patients. The standard measuring procedure involves manual segmentation of brain regions and the SI measurement, which are labor-intensive, time-consuming steps and subjective to the neuroradiologist. To help address this challenge, an automated signal intensity quantification software was developed and validated against manually obtained values from clinical routine patient data.
Full-text available
Valuable time and funding are spent on gathering fragmented and dispersedly stored data, and researchers are distracted from their main focus: understanding the brain and how disease affects it, and developing new neuroimaging tools to help in this endeavour. Data organization and solid analytics tools are essential to generate accurate and reproducible results. However, neuroscientists often face a critical problem hindering collaboration and reproducible research: lack of an efficient and standardized way to share data and to share analytic tools. We propose a web-based cloud system CloudNdesigned for neuroimaging workflow, enabling storage, quality control, version control, sharing, analysis, and visualization of various aspects of the neuroimaging data.
Full-text available
We introduce the Mindboggle-101 dataset, the largest and most complete set of free, publicly accessible, manually labeled human brain images. To manually label the macroscopic anatomy in magnetic resonance images of 101 healthy participants, we created a new cortical labeling protocol that relies on robust anatomical landmarks and minimal manual edits after initialization with automated labels. The "Desikan-Killiany-Tourville" (DKT) protocol is intended to improve the ease, consistency, and accuracy of labeling human cortical areas. Given how difficult it is to label brains, the Mindboggle-101 dataset is intended to serve as brain atlases for use in labeling other brains, as a normative dataset to establish morphometric variation in a healthy population for comparison against clinical populations, and contribute to the development, training, testing, and evaluation of automated registration and labeling algorithms. To this end, we also introduce benchmarks for the evaluation of such algorithms by comparing our manual labels with labels automatically generated by probabilistic and multi-atlas registration-based approaches. All data and related software and updated information are available on the website.
Background and purpose: In view of the recent observations that gadolinium deposits in brain tissue after intravenous injection, our aim of this study was to compare signal changes in the globus pallidus and dentate nucleus on unenhanced T1-weighted MR images in patients receiving serial doses of gadobutrol, a macrocyclic gadolinium-based contrast agent, with those seen in patients receiving linear gadolinium-based contrast agents. Materials and methods: This was a retrospective analysis of on-site patients with brain tumors. Fifty-nine patients received only gadobutrol, and 60 patients received only linear gadolinium-based contrast agents. Linear gadolinium-based contrast agents included gadoversetamide, gadobenate dimeglumine, and gadodiamide. T1 signal intensity in the globus pallidus, dentate nucleus, and pons was measured on the precontrast portions of patients' first and seventh brain MRIs. Ratios of signal intensity comparing the globus pallidus with the pons (globus pallidus/pons) and dentate nucleus with the pons (dentate nucleus/pons) were calculated. Changes in the above signal intensity ratios were compared within the gadobutrol and linear agent groups, as well as between groups. Results: The dentate nucleus/pons signal ratio increased in the linear gadolinium-based contrast agent group (t = 4.215, P < .001), while no significant increase was seen in the gadobutrol group (t = -1.422, P = .08). The globus pallidus/pons ratios followed similarly, with an increase in the linear gadolinium-based contrast agent group (t = 2.931, P < .0001) and no significant change in those receiving gadobutrol (t = 0.684, P = .25). Conclusions: Successive doses of gadobutrol do not result in T1 shortening compared with changes seen in linear gadolinium-based contrast agents.
The deep cerebellar nuclei (DCN) are a key element of the cortico-cerebellar loop. Because of their small size and functional diversity, it is difficult to study them using magnetic resonance imaging (MRI). To overcome these difficulties, we present here three related methodological advances. First, we used susceptibility-weighted imaging (SWI) at a high-field strength (7T) to identify the dentate, globose, emboliform and fastigial nucleus in 23 human participants. Due to their high iron content, the DCN are visible as hypo-intensities. Secondly, we generated probabilistic maps of the deep cerebellar nuclei in MNI space using a number of common normalization techniques. These maps can serve as a guide to the average location of the DCN, and are integrated into an existing probabilistic atlas of the human cerebellum (Diedrichsen et al., 2009). The maps also quantify the variability of the anatomical location of the deep cerebellar nuclei after normalization. Our results indicate that existing normalization techniques do not provide satisfactory overlap to analyze the functional specialization within the DCN. We therefore thirdly propose a ROI-driven normalization technique that utilizes both information from a T1-weighted image and the hypo-intensity from a T2*-weighted or SWI image to ensure overlap of the nuclei. These techniques will promote the study of the functional specialization of subregions of the DCN using MRI.
With the increasing use of three-dimensional MRI techniques it is becoming necessary to explore automated techniques for locating pathology in the volume images. The suitability of a specific technique to locate and identify healthy tissues of the brain was examined as a first step toward eventually identifying pathology in images. This technique, called multispectral image segmentation, is based on the classification of tissue types in an image according to their characteristics in various spectral regions. The spectral regions chosen for this study were the hydrogen spin-lattice relaxation time T1, spin-spin relaxation time T2, and spin density, rho. Single-echo, spin-echo magnetic resonance images of axial slices through the brain at the level of the lateral ventricles were recorded on a 1.5 Tesla imager from 20 volunteers ranging in age from 17 to 72 years. These images were used to calculate the T1, T2, and rho images used for the classification. Tissue classification was performed by locating clusters of pixels in a three-dimensional T1(-1)-T2(-1)-rho histogram. Gray matter, white matter, cerebrospinal fluid, meninges, muscle, and adipose tissues were readily classified in magnetic resonance images of the volunteers with a single set of T1, T2, and rho values. Cluster characteristics, such as size, shape, and location, provided information on the imaging procedure and tissue characteristics.
One of the most challenging problems in modern neuroimaging is detailed characterization of neurodegeneration. Quantifying spatial and longitudinal atrophy patterns is an important component of this process. These spatiotemporal signals will aid in discriminating between related diseases, such as frontotemporal dementia (FTD) and Alzheimer's disease (AD), which manifest themselves in the same at-risk population. Here, we develop a novel symmetric image normalization method (SyN) for maximizing the cross-correlation within the space of diffeomorphic maps and provide the Euler-Lagrange equations necessary for this optimization. We then turn to a careful evaluation of our method. Our evaluation uses gold standard, human cortical segmentation to contrast SyN's performance with a related elastic method and with the standard ITK implementation of Thirion's Demons algorithm. The new method compares favorably with both approaches, in particular when the distance between the template brain and the target brain is large. We then report the correlation of volumes gained by algorithmic cortical labelings of FTD and control subjects with those gained by the manual rater. This comparison shows that, of the three methods tested, SyN's volume measurements are the most strongly correlated with volume measurements gained by expert labeling. This study indicates that SyN, with cross-correlation, is a reliable method for normalizing and making anatomical measurements in volumetric MRI of patients and at-risk elderly individuals.
Gadolinium deposition in the brain: summary of evidence and recommendations
  • Dr Vikas Gulani
  • M D Fernando Calamante
  • Phd Frank G Shellock
  • Phd Emanuel Kanal
  • M D Scott
  • B Reeder
  • Md
Dr Vikas Gulani, MD. Prof Fernando Calamante, PhD. Prof Frank G Shellock, PhD. Prof Emanuel Kanal, MD. Prof Scott B Reeder, MD et al. (2017) Gadolinium deposition in the brain: summary of evidence and recommendations DOI: 10.1016/S1474-4422(17)30158-8.